In our Special Topics analysis of face recognition
research, the work of Dr. Patrick J. Flynn ranks at #4 by
total cites and #3 by cites per paper, based on 29 papers
cited a total of 681 times. His work also ranks in the top
1% among scientists publishing in the field of Computer
Science in
Essential Science IndicatorsSMfrom
Thomson
Reuters, and he also has Highly Cited Papers in the
field of Engineering in the database.
Dr. Flynn hails from the University of Notre Dame, where he is
Professor of Computer Science and Engineering, and Concurrent Professor of
Electrical Engineering. He is also an Associate Editor of the journals
IEEE Transactions on Image Processing and IEEE Transactions on
Information Forensics and Security. He is also the Vice President for
Conferences of the IEEE Biometrics Council.
Below, he talks with
ScienceWatch.com about his highly cited work as it relates
to face recognition.
Would you tell us a bit about your
educational background and research experiences?
I have a B.S. degree in Electrical Engineering and an M.S. and Ph.D. in
Computer Science, all from Michigan State University. I've been a professor
for 19 years, and worked at three different universities during that time.
My research work began as an undergraduate (helping to design systems to
detect vital signs from a distance), and my graduate studies focused on
computer vision and pattern recognition. Work early in my career involved
multiple extensions to my thesis work on object recognition as well as
projects in ultrasound imaging, computer graphics and visualization, and
image coding/compression.
How did you get involved in facial recognition
research?
I began to work in biometrics in the summer of 2001 with Kevin Bowyer,
co-director of my research group here at Notre Dame. Over the years, we
have established a well-respected group that has made good contributions to
the advances in biometrics research and technology. Our basic research
activity is supplemented by a long-term effort to collect biometric samples
from consenting subjects using an IRB-approved protocol.
"The research community has gained a broader and deeper
understanding of the challenges to effective iris image
matching and is now actively pushing to expand the envelope
of viable recognition to include recognition of persons at
some distance from the sensor, images with visible light
instead of infrared illumination, and combining iris
matching with other modes."
Over the last seven years, the group has collected in excess of 200,000
images and videos of faces, irises, and other sites imaged with a variety
of sensors. Much of the data is made available to other research groups and
companies working in the field, often as part of US government-sponsored
evaluations of biometrics technology.
One of your highly cited papers in our analysis is
the 2006 Computer Vision and Image Understanding paper, "A
survey of approaches and challenges in 3D and multi-modal 3D+2D face
recognition" (Bowyer KW, Chang K, Flynn P, 101[1]: 1-15). Would you
walk our readers through this paper and why you think it is garnering
citations?
This paper surveys the state of the art, circa 2006, in face recognition
using 3D shape instead of, or in addition to, 2D face photographs.
Three-dimensional sensors of various types have been around for several
decades, and there has been understandable interest to see whether human
identification from 3D face shape is superior to, comparable to, or
inferior to identification from 2D photographs. 3D face images are less
prone to strong contamination by lighting changes than 2D face images, and
face pose (head position and orientation) can be standardized using a 3D
model. Both of these advantages can be exploited by face recognition
systems.
The paper lists 25 systems that recognize faces from 3D information alone,
and provides summary descriptions of the techniques. It also lists 11
systems that combine 3D and 2D face images to perform recognition. It then
concludes with a listing of trends or key issues in this research area,
including the needs for better 3D sensors, algorithms, and experimental
methodology.
A few of your recent papers involve iris
recognition—would you talk a bit about this aspect of your work?
The idea that machines can learn to detect gender from the iris is
particularly intriguing—how is that done?
Iris recognition research has seen explosive growth over the past several
years and our group has been able to contribute to this community of
researchers in two ways, namely the provision of iris image data sets and
our basic research results on aspects of iris recognition. The research
community has gained a broader and deeper understanding of the challenges
to effective iris image matching and is now actively pushing to expand the
envelope of viable recognition to include recognition of persons at some
distance from the sensor, images with visible light instead of infrared
illumination, and combining iris matching with other modes. Kevin Bowyer's
recent work on gender prediction from iris images began as an undergraduate
student project, and is an exploration of data-mining concepts applied to
geometric features extracted from the iris image.
What are your hopes for progress in facial
recognition research over the next decade?
Face recognition systems will continue to advance in capability. We have
seen the emergence of face recognition in the consumer market through
products such as Picasa from Google and iPhoto from Apple. In these
products, face recognizers are used to automate portions of the tiresome
task of photo album management. This software is clearly limited in scope,
thus concerns about privacy do not apply (indeed, the major criticisms of
these programs have related to poor performance on low-quality images).
Additional applications of face recognition may pop up in the consumer
space, and the research community may contribute directly to these.
Security applications will still continue to motivate a lot of face
recognition research, particularly recognition of people in non-ideal
situations where the imagery is uncontrolled or poor in quality in other
ways. Face recognition in surveillance video is of strong interest and has
also proven to be very challenging.
What would you like the "take-away lesson" about
your research to be?
The summary message is that there are numerous basic and applied science
problems in the general area of biometrics, and many real and potential
challenges to both high-quality performance and broad acceptance in
society. Researchers have an obligation to consider both of these issues,
and contribute in both areas if they can.
Patrick J. Flynn, Ph.D.
Department of Computer Science and Engineering
University of Notre Dame
Notre Dame, IN, USA